Tianzi Zang, Cheng Xu, Jiayi Sun, Mingyuan Gao, Yicong Li, Juan Li, Wenze Ma · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840861
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Federated recommendation is an emerging paradigm that enables collaborative model training while preserving user privacy by keeping raw interaction data on local devices. However, in this setting, each client performs local training solely on a single user's limited historical interactions, which further exacerbates the data sparsity issue that has long been a bottleneck limiting recommendation performance. In this paper, we delve into the data sparsity problem in federated recommendation and propose FedPerGen, a personalized generative augmentation framework for federated recommendation. Specifically, each client is required to upload a user preference proxy to the server. The server then feeds this proxy along with diversity-constrained random noise into a generative network to synthesize personalized pseudo-item embeddings for each client. To ensure the informativeness and diversity of the generated embeddings, we design three complementary optimization objectives: feature-contrastive loss, model-alignment loss, and embedding-diversity loss. Furthermore, a two-stage warm-up training strategy is proposed to guarantee reliable user preference proxies and high-quality pseudo embeddings. Extensive experiments on real-world datasets demonstrate that FedPerGen consistently improves various federated recommendation baselines. The code is available at https://github.com/xcheng202202/FedPerGen .
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.